enable skipping of SW attention layers when using FP8 KV cache (#33695)

Signed-off-by: Jonas Kuebler <kuebj@amazon.com>
This commit is contained in:
Jonas M. Kübler
2026-03-27 07:25:02 -06:00
committed by GitHub
parent b111f8a61f
commit 98e7f223b9
4 changed files with 58 additions and 0 deletions
+7
View File
@@ -597,6 +597,9 @@ class EngineArgs:
attention_backend: AttentionBackendEnum | None = AttentionConfig.backend
calculate_kv_scales: bool = CacheConfig.calculate_kv_scales
kv_cache_dtype_skip_layers: list[str] = get_field(
CacheConfig, "kv_cache_dtype_skip_layers"
)
mamba_cache_dtype: MambaDType = CacheConfig.mamba_cache_dtype
mamba_ssm_cache_dtype: MambaDType = CacheConfig.mamba_ssm_cache_dtype
mamba_block_size: int | None = get_field(CacheConfig, "mamba_block_size")
@@ -1003,6 +1006,9 @@ class EngineArgs:
cache_group.add_argument(
"--calculate-kv-scales", **cache_kwargs["calculate_kv_scales"]
)
cache_group.add_argument(
"--kv-cache-dtype-skip-layers", **cache_kwargs["kv_cache_dtype_skip_layers"]
)
cache_group.add_argument(
"--kv-sharing-fast-prefill", **cache_kwargs["kv_sharing_fast_prefill"]
)
@@ -1578,6 +1584,7 @@ class EngineArgs:
enable_prefix_caching=self.enable_prefix_caching,
prefix_caching_hash_algo=self.prefix_caching_hash_algo,
calculate_kv_scales=self.calculate_kv_scales,
kv_cache_dtype_skip_layers=self.kv_cache_dtype_skip_layers,
kv_sharing_fast_prefill=self.kv_sharing_fast_prefill,
mamba_cache_dtype=self.mamba_cache_dtype,
mamba_ssm_cache_dtype=self.mamba_ssm_cache_dtype,